%% Based on a TeXnicCenter-Template by Gyorgy SZEIDL.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%------------------------------------------------------------
%
\documentclass[a4paper, 12pt, onecolumn]{article}
%%Options -- Point size:  10pt (default), 11pt, 12pt
%        -- Paper size:  letterpaper (default), a4paper, a5paper, b5paper
%                        legalpaper, executivepaper
%        -- Orientation  (portrait is the default)
%                        landscape
%        -- Print size:  oneside (default), twoside
%        -- Quality      final(default), draft
%        -- Title page   notitlepage, titlepage(default)
%        -- Columns      onecolumn(default), twocolumn
%        -- Equation numbering (equation numbers on the right is the default)
%                        leqno
%        -- Displayed equations (centered is the default)
%                        fleqn (equations start at the same distance from the right side)
%        -- Open bibliography style (closed is the default)
%                        openbib
% For instance the command
%           \documentclass[a4paper,12pt,leqno]{article}
% ensures that the paper size is a4, the fonts are typeset at the size 12p
% and the equation numbers are on the left side

\usepackage[left=25mm, right=25mm, top=25mm, bottom=25mm]{geometry}

\usepackage{pdflscape}
\usepackage{amsmath}
\usepackage{afterpage}
\usepackage{epigraph}
\setlength\epigraphwidth{.8\textwidth}
\allowdisplaybreaks[4]

\usepackage[utf8]{inputenc}

\usepackage[maxbibnames=5,maxcitenames=2,citestyle=authoryear,uniquename=mininit,bibstyle=authoryear,natbib=true,autocite=inline,sortcites=true,block=space,isbn=false,url=false,doi=false, backend = biber]{biblatex}

% Define commands/macros for paths that are used so we can switch quickly from the folder structure we work withto the flat folder structure required for submission.
\newcommand{\plotsdir}{../CovidIdeologyReplication/generated_images}
\newcommand{\texdir}{../CovidIdeologyReplication/generated_tex}

\bibliography{bibliography.bib}
\input{psrm_styles.tex}

\usepackage{titling}
\usepackage{verbatim}
\usepackage{comment}
\usepackage{dcolumn}
%\usepackage{epstopdf}
\usepackage{setspace}
\doublespacing
%\onehalfspacing
\usepackage[pdftex]{graphicx}
%\usepackage{subfig}
\newtheorem{hypothesis}{Hypothesis}
\newtheorem{treatment}{Treatment}
\usepackage{tikz}
%\usepackage{color,soul}
\allowdisplaybreaks[4]
\usepackage{longtable}
\usepackage{placeins}
\usepackage{booktabs}
\usepackage{amssymb}
\usepackage{subcaption}


\usepackage{array}
\newcolumntype{L}[1]{>{\raggedright\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}
\newcolumntype{C}[1]{>{\centering\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}
\newcolumntype{R}[1]{>{\raggedleft\let\newline\\\arraybackslash\hspace{0pt}}m{#1}}

% From https://tex.stackexchange.com/questions/18017/space-inserted-after-input-call
\newcommand{\inputy}[1]{\input{#1}\unskip}


\usepackage[T1]{fontenc}

\usepackage[%
colorlinks=true,linkcolor=black,citecolor=black,urlcolor=black,%
pdfstartview=FitH,%
bookmarksopen=true,bookmarksopenlevel=5,%
plainpages=false,pdfpagelabels,%
pagebackref=false]{hyperref}
\usepackage{cleveref}

\usepackage{xr}
\externaldocument{CovidIdeology}

% http://tex.stackexchange.com/questions/165128/how-to-stop-texcount-from-counting-begincomment-endcomment
%TC:envir comment [] ignore


\begin{document}
	
	\begin{titlingpage}
		%\setlength{\droptitle}{50pt} % lower the title
		
		
		
		\newcommand{\thetitle}{Supplementary Material for:\\
			The Non-Consequences of COVID-19 on Left-Right Ideological Beliefs}
		
		\title{\thetitle{}}
		\author{Jack Blumenau\\
			School of Public Policy,\\
			University College London\\
			\texttt{j.blumenau@ucl.ac.uk}\\
			\url{http://www.jackblumenau.com/}\\
			\and
			Timothy Hicks\\
			School of Public Policy,\\
			University College London\\
			\texttt{t.hicks@ucl.ac.uk}\\
			\url{http://tim.hicks.me.uk/}\\
			\and
			Alan M. Jacobs\\
			Department of Political Science,\\
			University of British Columbia\\
			\texttt{alan.jacobs@ubc.ca}\\
			\url{https://politics.ubc.ca/profile/alan-jacobs/}\\
			\and
			J. Scott Matthews\\
			Department of Political Science,\\
			Memorial University of Newfoundland\\
			\texttt{scott.matthews@mun.ca}\\
			\url{https://sites.google.com/view/jsmatthews/home}\\
			\and
			Tom O'Grady\\
			School of Public Policy,\\
			University College London\\
			\texttt{t.o'grady@ucl.ac.uk}\\
			\url{https://tomogradypolitics.wordpress.com/}
		}
		
		\date{Produced: \today}
		\normalsize
		
		\begin{onehalfspace}
			\maketitle
		\end{onehalfspace}
		
		
	\end{titlingpage}
	
	\begin{center}
		\begin{bfseries}
			\thetitle{}
		\end{bfseries}
		$ $\\
		%Word count: XXX
	\end{center}
	
	\setcounter{table}{0}
	\renewcommand{\thetable}{A\arabic{table}}
	\setcounter{figure}{0}
	\renewcommand{\thefigure}{A\arabic{figure}}
	\setcounter{equation}{0}
	\renewcommand{\theequation}{A\arabic{equation}}
	\appendix
	\renewcommand*{\thesection}{A\arabic{section}}
	
	%\section{Appendix}


\clearpage
\section{Question Wordings from Panel Data}
\label{apdx:PanelQWordings}

\begin{table}[htp]
	\centering
	\footnotesize
	\begin{tabular}{p{4.5cm} p{3.3cm} p{5cm} p{2.3cm}}
		\hline
		Survey Question  & Response Options  & Variable	& Range \\
		\hline
		It is the government's responsibility to provide a job for everyone who wants one & 1 Strongly disagree; \newline 2 Disagree; \newline 3 Neither/nor; \newline 4 Agree; \newline 5 Strongly agree; \newline 6 Don’t know & jobsForAll &  1--5 (No DK)\\
		When someone is unemployed, it's usually through no fault of their own	& As above	& reasonForUnemployment	& As above	\\
		
		Many people who get benefits don’t really deserve help	& As above	& benefitsNotDeserved	& As above	\\
		Some people feel that government should make much greater efforts to make people's incomes more equal. Other people feel that government should be much less concerned about how equal people's incomes are. Where would you place yourself and the political parties on this scale?	& 0 Government should try to make incomes equal;\newline 1;\newline \ldots \newline 9;\newline 10 Government should be less concerned about equal incomes;\newline DK	& redistSelf	& 0--10 (No DK)	\\
		Using the 0 to 10 scale below, where the end marked 0 means that government should cut taxes a lot and spend much less on health and social services, and the end marked 10 means that government should raise taxes a lot and spend much more on health and social services, where would you place yourself and the political parties on this scale? & As above& taxSpendSelf	& 0--10 (No DK)	\\
		During the next 12 months, how likely or unlikely is it that [y]ou will be out of a job and looking for work & 1 Very unlikely;\newline 2 Fairly unlikely;\newline 3 Neither/nor;\newline 4 Fairly likely;\newline 5 Very likely;\newline 99 Don't know	& riskUnemployment	& 1--5 (No DK)	\\
		During the next 12 months, how likely or unlikely is it that  [t]here will be times when you don't have enough money to cover your day to day living costs	& As above	& riskPoverty	& As above	\\
		\hline
	\end{tabular}
	\caption*{Table A1: Survey Questions}
	%\caption{Ideological dependent variables (part 1).}
	\label{tab:IdeologyDVs1}
\end{table}
\newpage


\clearpage
\section{Dependent Variable Survey Wave Coverage}
\label{apdx:WaveCoverage}

\begin{figure}[htp]
	\centering
	\includegraphics[width=\textwidth]{\plotsdir/DV_by_wave.png}
	\caption{Dependent variable coverage by wave}
	\label{fig:dv_by_wave}
\end{figure}



\section{Further Information for the Survey Experiment}
\label{apdx:Exp}

\begin{table}[htp]
	\centering
	\footnotesize
	\begin{tabular}{c p{2.5cm} p{1.6cm} p{1.6cm} p{2.5cm} p{1.3cm} c}
		\hline
		Group	& Block 1							& Block 2								& Block 3								& Block 4		& Block 5	& Probability \\ 
		\hline
		A			& Pre-treatment battery	& Distractors & -						& Post-treatment battery	& PCE		& 2/12	\\
		B			& Pre-treatment battery	& Distractors & PCE				& Post-treatment battery	& -				& 2/12	\\ 
		C			& Pre-treatment battery	& Distractors & CRIP				& Post-treatment battery	& PCE		& 1/12	\\ 
		D			& Pre-treatment battery	& Distractors & PCE, CRIP		& Post-treatment battery	& -				& 1/12	\\ 
		E			& Pre-treatment battery	& Distractors & URIP				& Post-treatment battery	& PCE		& 1/12	\\ 
		F			& Pre-treatment battery	& Distractors & PCE, URIP		& Post-treatment battery	& -				& 1/12	\\ 
		G			& Pre-treatment battery	& Distractors & SCP				& Post-treatment battery	& PCE		& 1/12	\\ 
		H			& Pre-treatment battery	& Distractors & PCE, SCP		& Post-treatment battery	& -				& 1/12	\\ 
		I			& Pre-treatment battery	& Distractors & SIP				& Post-treatment battery	& PCE		& 1/12	\\ 
		J			& Pre-treatment battery	& Distractors & PCE, SIP		& Post-treatment battery	& -				& 1/12	\\ 
		\hline
	\end{tabular} 
	\caption{Structure of the survey experiment in terms of the sequencing of core questions and treatments across different experimental groups.}
	\label{tab:SurveyProtocol}
\end{table}


\subsection{Primes from the Survey Experiment}
\label{apdx:Exp:Primes}

The personal/close experience (PCE) of COVID-19 priming treatment is formed of the following questions:
\begin{treatment}{Personal/Close Experience (PCE)\\}
	We will now ask you some questions about the experiences of you and people you know during the coronavirus pandemic.
	
	Have you personally, or has someone you know (such as family, friends, neighbours, or coworkers), experienced any of the following kinds of financial loss as a result of the coronavirus pandemic? (Check all that apply)
	\begin{table}[h!]
		\centering
		\begin{tabular}{p{8cm} c c}
			\hline
			& You personally	& Someone you know \\
			\hline
			Lost your/their job	& X	& X	\\
			Lost some of your/their wages or salary	& X	& X	\\
			Lost some of your/their business or self-employment income	& X	& X	\\
			Saw your/their business fail	& X	& X	\\
			None of the above	& X	& X	\\
			\hline
		\end{tabular}
	\end{table}
	
	Have you personally, or has someone you know (such as family, friends, neighbours, or coworkers), received any of the following forms of government financial support as a result of the coronavirus outbreak? (Check all that apply)
	\begin{table}[h!]
		\centering
		\begin{tabular}{p{8cm} c c}
			\hline
			& You personally	& Someone you know \\
			\hline
			Coronavirus Job Retention Scheme (for furloughed employees)	& X	& X	\\
			Coronavirus Self-Employment Income Support Scheme	& X	& X	\\
			Universal credit	& X	& X	\\
			Other state benefits	& X	& X	\\
			None of the above	& X	& X	\\
			\hline
		\end{tabular}
	\end{table}
\end{treatment}


Our various ideological-link treatments all begin with the following text:
\begin{quote}
	The coronavirus pandemic has prompted debate about whether the UK government should play a different role in the economy going forward. We will now show you an example of an argument that has been put forward in this debate. Please read this argument carefully.
\end{quote}
\noindent All the treatments conclude with the following question, which we ask to try to create more cognitive engagement with the respective treatments.
\begin{quotation}
	How much do you agree or disagree with the argument above?
	\begin{itemize}
		\item Strongly disagree;
		\item  Disagree;
		\item Neither agree nor disagree;
		\item Agree;
		\item Strongly disagree;
		\item Don't know.
	\end{itemize}
\end{quotation}

The individual ideological-link treatments are given below. Note that the indicated bold formatting is included in the actual survey.

\begin{treatment}{State Incapacity Frame (SIP)\\}
	The devastating experience of {\bfseries the pandemic has shown us the failure of an idea}: of the notion {\bfseries that government should step back and let the market solve our problems}. This is an idea that’s proved incapable of providing security for Britons and that left the country unprepared when we were tested most. 
	
	Our {\bfseries care homes} are perhaps the clearest example of this. But we see the same tragic story in overstretched {\bfseries hospitals and GP surgeries}, in a {\bfseries testing-and-tracing system} that practically collapsed when we needed it most, in {\bfseries schools with ever-growing class sizes}, in our once proud town centres and high streets, in an {\bfseries economy so insecure} that millions of people can’t afford to isolate. 
	
	This must now be a moment to think again about the country that we want and to recognize the value of public services. {\bfseries The pandemic has shown what can go wrong when we do not let the state play its proper role in society}. We need a state that invests in British skills, science, universities and manufacturing; that provides world-class education for all of our children; and that can ensure people don’t have to leave their home town to have a chance of getting a good job and won’t leave university with crippling debt.
\end{treatment}

\begin{treatment}{State Capacity Frame (SCP)\\}
	As we emerge from the pandemic, we have seen the success of an idea: of the notion that {\bfseries government can solve problems that the market cannot}. This is an idea that’s proved capable of providing security for Britons and that helped the country succeed when we were tested most. 
	
	The {\bfseries rapid development and rollout of an effective COVID-19 vaccine} is perhaps the clearest example of this. But we see the same inspiring story in the {\bfseries performance of the NHS} through a period of unparalleled crisis, in the {\bfseries furlough scheme} that allowed millions of workers to stay safe at home while continuing to collect a paycheck, and in a {\bfseries massive testing programme} that allowed hundreds of thousands to get free COVID-19 tests every day. 
	
	This is a moment to think again about the country that we want and to recognize the value of public services. {\bfseries The pandemic has shown what we can achieve when we let the state play its proper role in society}. We need a state that invests in British skills, science, universities and manufacturing; that provides world-class education for all of our children; and that can ensure people don’t have to leave their home town to have a chance of getting a good job and won’t leave university with crippling debt.
\end{treatment}

\begin{treatment}{Unequal Risk and Insurance Prime (URIP)\\}
	{\bfseries Covid has shown us the best of Britain, but it’s shown our fragilities too}. This virus has a deadly ability to find the most vulnerable and to expose deep inequalities and injustices in our society. 
	
	Tragically, this pandemic and economic crisis have shown that {\bfseries those who live in low-quality overcrowded housing}, who are {\bfseries trapped in insecure work}, and who {\bfseries live from paycheck-to-paycheck} can face financial catastrophe at any moment. We have seen that so many Britons are at risk of severe hardship through no fault of their own.
	
	Before the pandemic, we lived through a decade of neglect of our social safety net. {\bfseries We now need to seize this moment to build stronger benefits and national insurance schemes} and make sure that they can protect the most vulnerable Britons from risks beyond their control.
\end{treatment}

\begin{treatment}{Common Risk and Insurance Prime (CRIP)\\}
	{\bfseries Covid has shown us the best of Britain, but it’s shown our fragilities too}. This virus has a deadly ability to strike at every family and to expose how vulnerable all of us are.  
	
	Tragically, this pandemic and economic crisis have shown that {\bfseries even those who today live comfortably, have good jobs, and earn good wages can face financial catastrophe} at any moment. We have seen that so many Britons are at risk of severe hardship through no fault of their own. 
	
	Before the pandemic, we lived through a decade of neglect of our social safety net. {\bfseries We now need to seize this moment to build stronger benefits and national insurance schemes} and make sure that they can protect all Britons from risks beyond their control.
\end{treatment}


\clearpage
\subsection{Full Survey Experiment Question Wordings}
\label{apdx:Exp:QuestionWordings}

\subsubsection{redistSelf}

\begin{quotation}
	Some people feel that government should make much greater efforts to make people's incomes more equal. Other people feel that government should be much less concerned about how equal people's incomes are. Where would you place yourself and the political parties on this scale?
	\begin{itemize}
		\item 0 -- Government should try to make incomes equal
		\item 1
		\item \ldots
		\item 9
		\item 10 -- Government should be less concerned about equal incomes
		\item Don't know
	\end{itemize}
\end{quotation}

\subsubsection{taxSpendSelf}

\begin{quotation}
	Using the 0 to 10 scale below, where the end marked 0 means that government should cut taxes a lot and spend much less on health and social services, and the end marked 10 means that government should raise taxes a lot and spend much more on health and social services, where would you place yourself on this scale?
	\begin{itemize}
		\item 0 -- Government should cut taxes a lot and spend much less on health and social services
		\item 1
		\item \ldots
		\item 9
		\item 10 -- Government should increase taxes a lot and spend much more on health and social services
		\item Don't know
	\end{itemize}
\end{quotation}

\subsubsection{perceptionsOfPoorGrid}

\begin{quotation}
	How much do you agree or disagree with the following statements?
	%\begin{table}
	\centering
	\begin{footnotesize}
		\begin{tabular}{p{5.5cm} C{1.3cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm}}
			\hline
			Statement  & Strongly disagree & Disagree & Neither / nor & Agree & Strongly agree  & Don't know \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $reasonForUnemployment$]} \\
			When someone is unemployed, it's usually through no fault of their own &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $govtHandouts$]} \\
			Too many people these days like to rely on government handouts &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $benefitsNotDeserved$]} \\
			Many people who get benefits don’t really deserve help &  &  &  &  &  &  \\
			\hline
		\end{tabular}
	\end{footnotesize}
	%\end{table}
\end{quotation}


\subsubsection{bhpsRoGGrid}

\begin{quotation}
	People have different views about society and the economy. How much do you agree or disagree with the following statements?
	%\begin{table}
	\centering
	\begin{footnotesize}
		\begin{tabular}{p{5.5cm} C{1.3cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm}}
			\hline
			Statement  & Strongly disagree & Disagree & Neither / nor & Agree & Strongly agree  & Don't know \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $privateEnterprise$]} \\
			Private enterprise is the best way to solve Britain's economic problems &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $stateOwnership$]} \\
			Major public services and industries ought to be in state ownership &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $jobForAll$]} \\
			It is the government's responsibility to provide a job for everyone who wants one &  &  &  &  &  &  \\
			\hline
		\end{tabular}
	\end{footnotesize}
	%\end{table}
\end{quotation}

\subsubsection{isspRoGGrid}

\begin{quotation}
	Here are some things the government might do for the economy. Please show which actions you are in favour of and which you are against\ldots
	%\begin{table}
	\centering
	\begin{footnotesize}
		\begin{tabular}{p{5.5cm} C{1.3cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm}}
			\hline
			Statement  & Strongly in favour of & In favour of & Neither / nor & Against & Strongly against  & Don't know \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $financeNewJobs$]} \\
			Government financing of projects to create new jobs &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $supportNewProducts$]} \\
			Support for industry to develop new products and technology &  &  &  &  &  &  \\
			\hline
		\end{tabular}
	\end{footnotesize}
	%\end{table}
\end{quotation}


\subsubsection{bsasBenefitsGrid}

\begin{quotation}
	Some people think that there should be more government spending on social security, while other people disagree. For each of the groups I read out please say whether you would like to see more or less government spending on them than now. Bear in mind that if you want more spending, this would probably mean that you would have to pay more taxes. If you want less spending, this would probably mean paying less taxes. 
	%\begin{table}
	\centering
	\begin{footnotesize}
		\begin{tabular}{p{5.5cm} C{1.3cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm}}
			\hline
			Statement  & Spend much more & Spend more & Same as now & Spend less & Spend much less & Don't know \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $poorBenefitsMore$]} \\
			Benefits for the poor &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $unempBenefitsMore$]} \\
			Benefits for unemployed people &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $disabledBenefitsMore$]} \\
			Benefits for disabled people who cannot work &  &  &  &  &  &  \\
			\hline
		\end{tabular}
	\end{footnotesize}
	%\end{table}
\end{quotation}


\subsubsection{riskGrid}

\begin{quotation}
	During the next 12 months, how likely or unlikely is it that\ldots\\
	%\begin{table}
	\centering
	\begin{footnotesize}
		\begin{tabular}{p{5.5cm} C{1.3cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm} C{1.2cm}}
			\hline
			Statement  & Very unlikely & Fairly unlikely & Neither / nor & Fairly likely & Very likely & Don't know \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $riskPoverty$]} \\
			There will be times when you don't have enough money to cover your day to day living costs &  &  &  &  &  &  \\
			\hline
			\multicolumn{7}{l}{[Resulting variable: $riskUnemployment$]} \\
			You will be out of a job and looking for work &  &  &  &  &  &  \\
			\hline
		\end{tabular}
	\end{footnotesize}
	%\end{table}
\end{quotation}


\clearpage
\subsection{PCA Loadings}
\label{apdx:Exp:PCALoadings}

\begin{table}[!htbp] 
	\label{redistribution_pca_loadings} 
	\center
	\caption{Principal component analysis loadings for ``redistribution'' items } 
	\input{\texdir/redistribution_pca_loadings.tex}\\
	%\emph{Note:} 
\end{table}

\begin{table}[!htbp] 
	\label{rog_pca_loadings} 
	\center
	\caption{Principal component analysis loadings for ``role of government'' items } 
	\input{\texdir/rog_pca_loadings.tex}\\
	%\emph{Note:} 
\end{table}
\newpage

\clearpage
\section{Regression Tables}
\label{apdx:Exp:RegressionTables}
\FloatBarrier

\begin{table}[h]
	\centering
	\caption{Ideological effects of Crisis Exposure: Estimates of $\hat{\beta}_1$ and $\hat{\beta}_2$ for $redistSelf$, $taxSpendSelf$ and $jobsForAll$.}\label{tab:baseline_ideoresults}
	\input{\texdir/fe_baseline_ideo_vars.tex}
\end{table}


\begin{table}[h]
	\centering
	\caption{Ideological effects of Crisis Exposure: Estimates $CovidFurlough$, $CovidSelfEmployment$, $CovidUniversalCredit$ and $CovidLostEmployment$ for $redistSelf$, $taxSpendSelf$ and $jobsForAll$.}\label{tab:baseline_multi_treatment_ideo_vars}
	\input{\texdir/fe_multi_treatment_ideo_vars.tex}
\end{table}


\begin{table}[h]
	\centering
	\caption{Ideological effects of Crisis Exposure: Estimates $CovidFurlough$, $CovidSelfEmployment$, $CovidUniversalCredit$ and $CovidLostEmployment$ interacted with with party supported in 2019 for $redistSelf$, $taxSpendSelf$ and $jobsForAll$.}\label{tab:fe_by_party_by_treat}
	\input{\texdir/fe_by_party_by_treat.tex}
\end{table}


\begin{table}[h]
	\centering
	\caption{Ideological effects of Crisis Exposure: Estimates $CovidFurlough$, $CovidSelfEmployment$, $CovidUniversalCredit$ and $CovidLostEmployment$ for $govtHandouts$ and $reasonForUnemployment$.}\label{tab:fe_multi_treatment_deserving_vars}
	\input{\texdir/fe_multi_treatment_deserving_vars.tex}
\end{table}


\begin{table}[h]
	\centering
	\caption{Ideological effects of Crisis Exposure: Estimates of $\hat{\beta}_1$ and $\hat{\beta}_2$ for $riskPoverty$ and $riskUnemployment$.}\label{tab:baseline_risk_results}
	\input{\texdir/fe_baseline_risk_vars.tex}
\end{table}


\clearpage
\section{Differential Aggregate Pandemic Attitude Change By Pre-Pandemic Attitude}
\label{apdx:DiffAggByPrior}

Aggregate stability may, in principle, conceal significant change within particular subgroups. In this section, we consider whether those with different left-right ideological beliefs $prior$ to the pandemic responded differently to the economic crisis. We capture prior attitudes by averaging respondents' views -- separately for $redistSelf$ and $taxSpendSelf$ -- across all pre-crisis waves in which they appear. We group respondents into three categories according to where their average response falls on the 0--10 scale of the underlying measures: ``right'' (0 to 4), ``mid'' (>4 to 6), and ``left'' (>6 to 10). Figure~\ref{fig:taxSpendSelf_by_prior_by_wave_21} plots these data for $taxSpendSelf$ and figure~\ref{fig:redistSelf_by_prior_by_wave_21} does so for $redistSelf$.

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/taxSpendSelf_by_prior_by_wave_21.pdf}
	\caption{Through-time plot of average $taxSpendSelf$ values for respondents defined as ``right'', ``mid', or ``left'' based on their pre-pandemic $taxSpendSelf$ responses.}
	\label{fig:taxSpendSelf_by_prior_by_wave_21}
\end{figure}

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/redistSelf_by_prior_by_wave_21.pdf}
	\caption{Through-time plot of average $redistSelf$ values for respondents defined as ``right'', ``mid', or ``left'' based on their pre-pandemic $redistSelf$ responses.}
	\label{fig:redistSelf_by_prior_by_wave_21}
\end{figure}

On both measures, over-time change at the subgroup level is generally trivial in magnitude (though, owing to our very large sample size, some over-time differences are significant at conventional levels). The major exception occurs among those whose average pre-crisis attitude on $taxSpendSelf$ placed them on the ideological right: between the final pre-crisis and first post-crisis waves of our panel, this group shifts notably leftward on $taxSpendSelf$ -- a difference in average attitudes covering nearly a fifth of the measure's range. Given we do not observe a similar shift on $redistSelf$, we are sceptical that it reflects broad change in ideological beliefs; indeed, as $taxSpendSelf$ derives from a question that highlights spending on ``health care'', the dynamic may simply reflect a short-term rally to a National Health Service under very severe and salient strain. We also note that, on the $taxSpendSelf$ measure, the ``right'' are a small minority, including just over 10 percent of the available sample.


\clearpage
\section{Parallel Trends}
\label{apdx:Exp:ParallelTrends}

The key identifying assumption underpinning our generalised difference-in-differences design is that there are parallel trends in the outcome variables between treated and untreated units. In our context, this implies that respondents who were not directly affected by the COVID-19 pandemic (in terms of losing employment or receiving government support) provide a reasonable counterfactual for the trends in the outcomes for those who were directly affected. Although we cannot test this assumption directly, figure \ref{fig:parallel_trends} demonstrates the plausibility of this assumption by plotting the $redistSelf$ and $taxSpendSelf$ average outcomes for treatment and control groups defined by $CovidLostEmployment$ and $CovidGovSupport$ over time. 

Panels in the top row show the trends for treatment (red) and control (black) groups defined by the $CovidGovSupport$ treatment, which includes respondents who were beneficiaries of \emph{any} form of financial support from the UK government during the crisis (the furlough scheme, the self-employment income support scheme, or Universal Credit). Panels in the bottom row show the treatment and control trends defined by the $CovidLostEmployment$ treatment, which is equal to one for any respondent who reports that their working hours were reduced during the crisis period relative to before the crisis. All average outcomes in the plots are weighted by cross-sectional BES survey weights.

\afterpage{
	\begin{landscape}
		\begin{figure}[htp]
			\centering
			%	\hspace{-2cm}
			\includegraphics[width=1.25\textwidth]{\plotsdir/parallel_trends_redistSelf_taxSpendSelf.pdf}
			\caption{Over-time trends in $redistSelf$ and $taxSpendSelf$ for treatment and control groups defined by $CovidLostEmployment$ and $CovidGovSupport$. Points represent the mean outcome for treatment (red) and control (black) groups in each survey wave. Lines represent 95\% confidence intervals. All outcomes are weighted by cross-sectional BES survey weights. }
			\label{fig:parallel_trends}
		\end{figure}
	\end{landscape}
}

The averages plotted in the figure lend credibility to the parallel trends assumption. In the pre-crisis period, the treatment and control trends for either treatment for the $redistSelf$ outcome are very similar and, consistent with the null effects presented in the main body of the paper, we see essentially no divergence in these trends during the crisis. The trends for treatment groups defined by the $CovidLostEmployment$ variable and the $taxSpendSelf$ outcome are noisier, partly as a function of the limited number of pre-treatment waves we observe for this variable and partly because only a small fraction of our sample are coded as treated for this variable (see table \ref{tab:treat_dist_by_wave} in the main body of the paper). However, the treatment and control trends in $taxSpendSelf$ for the $CovidGovSupport$ variable are much closer to parallel, again supporting the identification assumption that underpins our analysis.


\clearpage
\section{`Visibility' Results from the Panel Analysis}
\label{apdx:Visibility}

Figure~\ref{fig:visibility_ideoresults} shows fixed-effects coefficients from a regression that omits $CovidGovSupport$ and instead includes $CovidFurlough$, $CovidSelfEmployment$ and $CovidUniversalCredit$. See table~\ref{tab:baseline_multi_treatment_ideo_vars} for full results.

\begin{figure}[htp]
	\centering
	%\hspace{-2cm}
	\includegraphics[width=1\textwidth]{\plotsdir/fe_multi_treatment_ideo_vars.pdf}
	\caption{Ideological effects of Crisis Exposure, Splitting out Different forms of Government Assistance (coefficients and 95\% Confidence Intervals)}
	\label{fig:visibility_ideoresults}
\end{figure}



\clearpage
\section{Evaluating Treatment Effects in the Medium Term}
\label{apdx:Exp:LongRun}

The data we use to estimate the fixed-effect analyses presented in the main body of the paper come from three post-crisis waves, fielded in April, June, and September of 2020. A reasonable question is whether these survey dates include enough time after the onset of the crisis for effects of crisis-exposure on political attitudes to become manifest. 

In order to address this question, we use additional data from the 21st wave of the BES panel, conducted in May 2021. Wave 21 included our three main ideological outcome variables, but did not include the questions that form the basis of our key treatment variables.  This means that we cannot replicate our full analyses for this later time period. However, all waves of the BES panel ask respondents to report their current working status, and in all four of the COVID-affected waves in our data ``furloughed'' was included as one of the response options to this question. We code a dummy variable, $FurloughSelf_{i,t}$, which is equal to 1 if a respondent $i$ reports being furloughed in time $t$, and 0 otherwise.

This question gives a narrower measure of crisis-exposure than the $CovidFurloughed$ variable we describe in the main body of the paper, which is coded 1 if the respondent reports that either they \emph{or a member of their household} were beneficiaries of the furlough scheme. However, the advantage of using this variable is that it allows us to include responses to wave 21 of the BES, which is from a later date in the pandemic.

We use this variable to estimate regressions which are similar to the baseline specifications described in equation \ref{eq:fe_model} of the main body of the paper. In particular, here we estimate the effects of employment loss and being a beneficiary of the furlough scheme using models of the form:

\begin{eqnarray}\label{eq:fe_model_w21}
	Y_{i,t}	&	=	&	\beta_1 FurloughSelf_{i,t} + \beta_2 CovidLostEmployment_{i,t}  \nonumber \\
	&		& + \alpha_i + \delta_t + \epsilon_{i,t}
\end{eqnarray}
\noindent where $Y_{i,t}$ is the outcome (i.e. $redistSelf$ or $taxSpendSelf$) for respondent $i$ in wave $t$, and $\alpha_i$ and $\delta_t$ are fixed-effects for respondent and survey wave, respectively. We estimate the model described in equation \ref{eq:fe_model_w21} twice: once using data up to and including the second PACER wave (September 2020), and once using data up to and including the 21st BES wave (May 2021). 

\begin{figure}[h]
	\centering
	%	\hspace{-2cm}
	\includegraphics[width=\textwidth]{\plotsdir/fe_baseline_ideo_vars_w21_comparison.pdf}
	\caption{The plot depicts the effects of $FurloughSelf_{i,t}$ on $redistSelf$ and $taxSpendSelf$. Estimates come from a regression of the outcomes on the furlough indicator, a dummy for whether the respondent lost employment due to COVID, and respondent and survey-wave fixed effects. Black points illustrate the effects estimated using data up to the second PACER wave (September 2020) and red points illustrate effects estimated using data up to BES wave 21 (May 2021). Lines represent 95\% confidence intervals. All regressions are weighted by cross-sectional BES survey weights. }
	\label{fig:ideo_effects_w21_comparison}
\end{figure}

Figure \ref{fig:ideo_effects_w21_comparison} depicts the effects of $FurloughSelf_{i,t}$ on $redistSelf$ and $taxSpendSelf$ ($\beta_1$ in equation \ref{eq:fe_model_w21}). Points in red represent estimates from data \emph{including} BES wave 21 and points in black represent estimates \emph{excluding} wave 21. There are two notable features of these results. First, the estimates in figure \ref{fig:ideo_effects_w21_comparison} are very similar to those presented in the figures in the main body of the paper, indicating that the narrower definition of the $FurloughSelf$ variable affect our conclusions about the (null) effects of government COVID-support on political attitudes. Second, and more importantly, the results are entirely insensitive to the inclusion of data from the later treatment period. That the effects are stable to the inclusion of data from May 2021 suggests that our conclusions are unlikely to be driven by the limited period of post-treatment data that we have available to us.

\afterpage{
	\begin{landscape}
		\begin{figure}[htp]
			\centering
			\vspace{2cm}
			\includegraphics[width=1.25\textwidth]{\plotsdir/parallel_trends_redistSelf_taxSpendSelf_w21_comparison.pdf}
			\caption{Over-time trends in $redistSelf$ and $taxSpendSelf$ for treatment and control groups defined by $CovidLostEmployment$ and $CovidGovSupport$. Points represent the mean outcome for treatment (red) and control (black) groups in each survey wave. Lines represent 95\% confidence intervals. All outcomes are weighted by cross-sectional BES survey weights. }
			\label{fig:parallel_trends_w21}
		\end{figure}
	\end{landscape}
}

Figure \ref{fig:parallel_trends_w21} replicates the parallel trends analysis depicted in figure \ref{fig:parallel_trends}, but here we focus on the trends for furloughed and not-furloughed individuals before and after the onset of the pandemic. As this analysis is based on the $FurloughSelf_{i,t}$ variable, we are able to include treatment-group averages for BES wave 21 respondents in the plot. There is very little difference between the wave 21 averages and the group averages for the other post-crisis-onset periods. This again provides reassuring evidence that the null effects we document in the paper are unlikely to be due to the limited time period that we study. 

In summary, even looking as far as 16 months after the onset of the pandemic in the UK, we find no evidence that those exposed to pandemic-induced financial shocks changed their political attitudes as a result of the crisis. 

\clearpage
\section{Subgroup Analyses}
\label{apdx:Exp:Subgroups}

In section~\ref{sec:limitedeffects:partyid} of the paper, we show that experience of the pandemic did not appear to differentially affect respondents with different ideological `starting points', as measured by the party for whom they voted in the 2019 general election. In this section, we investigate further subgroup-specific treatment effects, all of which engage with the idea that respondents' prior experiences or attitudes might condition their responses to personal experience of the pandemic. 

We focus on four main comparisons. First, we examine whether respondents with different \emph{prior attitudes} on our main outcome variables were differentially responsive to exposure to the pandemic. As described in section~\ref{sec:limitedeffects:partyid}, there are reasons to expect that respondents with more left-wing prior attitudes may be either more or less affected by personal experience of the crisis than respondents with more right-wing prior attitudes. To test these expectations, we identify the set of respondents for whom we have pre-crisis measures of attitudes and we calculate respondents' average (i.e. across-wave) pre-crisis attitudes on these variables and split respondents into ``left'', ``mid'' and ``right'' groups based on where these average responses fall in the original response distribution of those variables.\footnote{We code any respondent in the 0-4 range as ``Left'' (10,976 respondents for $redistSelf$ and 6382 respondents for $taxSpendSelf$), respondents with average pre-crisis values between 4 and 6 as ``Mid'' (7984 respondents for $redistSelf$ and 5445 respondents for $taxSpendSelf$), and those higher than 6 as ``Right'' (8058 respondents for $redistSelf$ and 1518 respondents for $taxSpendSelf$). We also tried alternative approaches to coding prior beliefs, including restricting the range of pre-crisis surveys over which we average to those in the year before the crisis, as well as simply taking a respondent's most recent pre-crisis answer to the outcome questions and using that as the basis for the categorisation. None of the results presented here are sensitive to these choices. Finally, we note that this analysis is not possible to conduct at all for the jobsForAll variable, as we have no pre-crisis data available for that measure.}

Second, we investigate whether respondents who have expressed ambivalence via a \emph{``don't know'' response} to our main outcome variables in the past were differently affected by experience of the crisis. Under the assumption that ``Don't know'' respondents do not hold well-defined attitudes on the set of issues addressed by our survey questions, we ought to expect larger treatment effects for these respondents than for those who have not expressed such ambivalence in the past. For each respondent for whom we have pre-crisis measures of our outcome variables, we code whether they selected ``Don’t know'' for the relevant variable in any of the pre-crisis waves. Of the respondents in our sample, 2355 gave a ``Don’t know'' response for the $taxSpendSelf$ variable at some point during the pre-crisis period, and 8519 gave a ``Don’t know'' response for the $redistSelf$ variable at some point before the crisis.

Third, we consider whether respondents with different levels of \emph{political attention} are differentially responsive to the crisis, on the basis that less politically attentive respondents are also likely to be less politically knowledgeable and therefore may be more prone to attitudinal change in the face of a crisis. The BES includes responses to a question asking respondents “How much attention do you generally pay to politics?”. This is an 11-category variable ranging from ``Pay no attention'' to ``Pay a great deal of attention''. We calculate the pre-crisis mean response on this variable for each respondent and then categorize respondents into ``Low'' and ``High'' attention groups according to the median response for this variable across respondents (which is 8 on the 11 category scale).

Finally, we consider whether respondents who have reported receiving \emph{previous financial support from the government} are more or less affected by personal exposure to the crisis. As we argue throughout the paper, a distinctive feature of the COVID-19 crisis was not just the scale of the shock, but that many of the people who were affected by the crisis had had very little experience of either labour-market insecurity or state support before the crisis occurred. As a consequence, we might expect that the crisis had different effects for those people receiving state support for the first time than for people who had been recipients of state support in the past. To investigate this idea we take advantage of the fact that, in two pre-crisis waves of the BES, respondents were asked whether they or their household received income from a number of sources, including Job Seeker’s Allowance/Unemployment Benefit; Income Support or family credit; Invalidity, sickness or disabled pension or benefits; and other state benefits. We use this variable to code respondents into two groups according to whether they report having received any such government support in either of the two survey waves in which that question was asked.\footnote{This analysis is limited for two reasons. First, as these questions were asked in only two pre-crisis waves, there is a large amount of missingness in this data. Of the 32,352 respondents in our full sample, we have information on prior government support for only 13,890 respondents. Second, this measure captures only receipt of state support at given points in time (i.e. at the time of the two survey waves). Accordingly, it is possible that some individuals who report receiving no state support would have received state support at other times in the past, which is likely to cause a downward bias in our estimates of treatment-effect heterogeneity.}

We therefore have four variables -- left, mid or right prior attitudes; high versus low attention to politics; don't know respondents versus others; and prior government support recipients versus others -- which we use to look for heterogeneous treatment effects. We use these variables in fixed-effect regression models for the $redistSelf$ and $taxSpendSelf$ outcome variables of the following form: 
\begin{eqnarray}\label{eq:fe_model_interaction}
	Y_{i,t}	&	=	&	\beta_1 CovidFurlough_{i,t} + \nonumber \\
	&& \sum_{g=2}^G \beta_1^g (CovidFurlough_{i,t} \cdot Group_{i(g)}) +  \nonumber \\	
	&& \beta_2 CovidSelfEmployment_{i,t} + \nonumber \\
	&& \sum_{g=2}^G \beta_2^g (CovidSelfEmployment_{i,t} \cdot Group_{i(g)}) +  \nonumber \\	
	&& \beta_3 CovidUniversalCredit_{i,t} + \nonumber \\
	&& \sum_{g=2}^G \beta_3^g (CovidUniversalCredit_{i,t} \cdot Group_{i(g)}) +  \nonumber \\	
	&& \beta_4 CovidLostEmployment_{i,t} + \nonumber \\
	&& \sum_{g=2}^G \beta_4^g (CovidLostEmployment_{i,t} \cdot Group_{i(g)})  +  \nonumber \\	
	&& + \alpha_i + \delta_{t,g} + \epsilon_{i,t}
\end{eqnarray}
\noindent where $Y_{i,t}$ is the outcome (i.e. $redistSelf$ or $taxSpendSelf$) for respondent $i$ in wave $t$, and $\alpha_i$ and $\delta_{t,g}$ are fixed-effects for respondent and survey-wave-by-group, respectively. $Group_i$ represents one of the four grouping variables described above. The coefficients associated with the constituent treatment terms -- $CovidFurlough_{i,t}$, $CovidSelfEmployment_{i,t} $, $CovidUniversalCredit_{i,t}$, $CovidLostEmployment_{i,t}$ -- describe the treatment effects of each type of crisis exposure for respondents in the ``low'' group defined by the binary $Group$ variables. The treatment effects for those in other groups is given by the sum of the constituent term and the relevant interaction term (i.e. $\beta_4 + \beta_4^2$ is the effect of the lost employment treatment for respondents in the second group of the grouping variable). We present the full set of conditional treatment effects from these models in figures \ref{fig:subgroups_prior_attitudes}, \ref{fig:subgroups_dk}, \ref{fig:subgroups_attention} and \ref{fig:subgroups_prior_gov_support} below.

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/fe_by_prior_by_treat.pdf}
	\caption{Effects of crisis exposure on attitudes for respondents with different prior attitudes (coefficients and 95\% confidence intervals)}
	\label{fig:subgroups_prior_attitudes}
\end{figure}

Figure \ref{fig:subgroups_prior_attitudes} shows the estimated conditional treatment effects for respondents who gave left-, mid-, and right-leaning responses for the $redistSelf$ and $taxSpendSelf$ in the pre-crisis waves. As is evident, using this strategy, we find very limited evidence of treatment heterogeneity for either variable. There is some evidence that left-leaning respondents who received Universal Credit during the crisis became somewhat less supportive of higher tax and spend policies compared to left-leaning respondents who did not receive Universal Credit, but this effect is relatively small. In no other instance do we find evidence of significant heterogeneity by prior attitudes.

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/fe_by_prior_dk_by_treat.pdf}
	\caption{Effects of crisis exposure on attitudes for respondents with and without `Don't know' responses in the past (coefficients and 95\% confidence intervals)}
	\label{fig:subgroups_dk}
\end{figure}

Figure \ref{fig:subgroups_dk} shows the estimated conditional treatment effects for respondents who gave ``Don't know'' responses at some point before the crisis as well as for respondents who never gave a ``Don't know'' response. We again find only very limited evidence of treatment heterogeneity here. Although the point estimates differ for some treatments between ``Don't know'' and other respondents, these differences point in different directions across the treatment variables and outcomes. We are also not able to reject the null hypothesis for any of these interactions. In general, we conclude from this analysis that ``Don't know'' respondents were not systematically more responsive to experience of the crisis than other respondent types.


\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/fe_by_attention_by_treat.pdf}
	\caption{Effects of crisis exposure on attitudes for respondents with high and low levels of political attention (coefficients and 95\% confidence intervals)}
	\label{fig:subgroups_attention}
\end{figure}

Figure \ref{fig:subgroups_attention} shows the estimated conditional treatment effects for high and low attention respondents and reveals that we again find very limited evidence of treatment heterogeneity. The treatment effect point estimates again differ in small ways between low and high attention respondents, but these differences point in different directions across the treatment variables and outcomes. Again we are only able to reject the null hypothesis for the interaction on the furlough treatment variable on the $redistSelf$ outcome, where we estimate that high-attention respondents report attitudes that are more pro-redistribution during the pandemic than they did before the crisis (and there is no effect for low-attention respondents). In general, we conclude from this analysis that respondents with high and low levels of political attention did not have systematically different responses to exposure to the crisis.


\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/fe_by_prior_gov_support_by_treat.pdf}
	\caption{Effects of crisis exposure on attitudes for respondents who have, and have not, received government support in the past (coefficients and 95\% confidence intervals)}
	\label{fig:subgroups_prior_gov_support}
\end{figure}

Finally, figure \ref{fig:subgroups_prior_gov_support} compares the estimated treatment effects for respondents who reported receiving some form of government support in the past to the same quantities for respondents who did not report receiving such support. We find no evidence of treatment heterogeneity with respect to this decomposition of our respondents: we are unable to reject the null hypothesis of no difference for any of the interactions in equation \ref{eq:fe_model_interaction}. In general, we conclude from this analysis that exposure to the crisis -- either via job loss or receipt of government support -- was largely uniform across respondents, and was not significantly affected by respondents’ historical receipt of state support.


\clearpage
\section{Job loss and benefit receipt}
\label{apdx:JobLossBenefitInteraction}

In this section, we evaluate whether the effects of job loss on attitudes are conditional on whether the respondent was also a recipient of (non-pandemic-specific) state benefits. There are various plausible ways in which job loss and benefit receipt might interact to affect attitudes. For instance, it might be that people who suffer an employment shock shift their attitudes to the left in response, but that these effects are mitigated by receiving benefits. By contrast, it could be that benefit receipt -- by making social insurance schemes more visible -- amplify the effects of job loss on attitudes. In either case, we would expect the effect of losing employment on attitudes to vary according to whether an individual also receives some kind of unemployment support from government.

Table \ref{tab:fe_multi_treatment_employment_uc_interaction} reports the results of an analysis which replicates the analysis presented in figure~\ref{fig:visibility_ideoresults}, but where we additionally interact the $CovidLostEmployment$ and $CovidUniversalCredit$ variables. If the effects of employment loss on attitudes are conditional on the receipt of state benefits, then the interaction on this variable will differ from zero.

\begin{table}[htp]
	\centering
	\caption{Ideological effects of Crisis Exposure: Interacting Employment Loss and Receipt of Universal Credit}\label{tab:fe_multi_treatment_employment_uc_interaction}
	{\footnotesize
		\input{\texdir/fe_multi_treatment_employment_uc_interaction.tex}
	}
\end{table}

The table shows that we find mixed evidence for this conditional relationship across our various outcome variables. While receiving state benefits appears to have no moderating effect for the $redistSelf$, $jobsForAll$, or $riskPoverty$ outcomes, we estimate significant interaction effects for the $taxSpendSelf$ and $riskUnemployment$ outcomes. For $taxSpendSelf$, losing employment pushes respondents slightly rightward when those respondents did not receive Universal Credit during the pandemic. By contrast, for those who \emph{did} receive Universal Credit, the effect of losing employment is to shift them moderately leftwards on the tax-spend question. A similar pattern can be seen for the $riskUnemployment$ variable, where the combination of losing employment \emph{and} receiving Universal Credit leads people to perceive a higher risk of unemployment in the future, relative to those who lost employment during the crisis but did not receive benefits.

While we see these patterns as potentially interesting, we caution against over-interpretation. In particular, we are at risk of slicing the data too thinly to draw robust inferences here. Of the 32,000 or so respondents that appear in our full sample, only 1002 reported losing employment during the pandemic, and of these only 170 also report receiving Universal Credit during the same period. We therefore have very few observations to inform the estimates of the interaction effects in the table, meaning that the documented effects are likely to be sensitive to sampling variability (a problem that is exacerbated by the fact that we are also testing multiple hypotheses across different outcome variables). 


\clearpage
\section{Evidence on Deservingness Perceptions from the British Social Attitudes Survey}
\label{apdx:BSAS}

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=0.9\textwidth]{\plotsdir/bsas_deservingness.pdf}
	\caption{Trends in Average Deservingness Perceptions in Great Britain, 2010-20 (British Social Attitudes Survey)}
	\label{fig:bsas_deservingness}
\end{figure}

\noindent Figure~\ref{fig:bsas_deservingness} shows data from the British Social Attitudes Survey on deservingness perceptions from 2010--20. These are annual cross-sectional surveys, with survey weights applied. Both questions ask respondents to agree or disagree with statements that paint welfare recipients as lazy, undeserving and fraudulent. In both cases the statements have 5-point likert response scales. Like the questions in the main paper, these are coded so that higher responses indicate more left-wing opinions: greater disagreement with the statements. They show that from 2012 onwards there was a continuous leftward shift in opinion. On average, British people felt that claimants were much more deserving in 2020 than in 2010--11, which \textcite{OGrady2022} attributes to a long-term softening of media and political discourse towards the British welfare system and its users. However, the leftward trend did not continue from 2019--20 over the course of the pandemic (2020 fieldwork was carried out in October-December, during the pandemic). Instead there was no aggregate change over the pandemic. This suggests that much of the leftward shift in the variable that featured in our survey ($govtHandouts$) probably also took place prior to the pandemic. Between the two sets of surveys there is little evidence of a pandemic-induced leftward shift in opinions about deservingness. 


\clearpage
\section{Experimental treatment effects and inferences}
\label{apdx:Exp:Results}

The explicit test of the moderation hypothesis given in equation~\ref{hyp:DeltaInteraction} in the main text is the combination of the following conditions from the estimated equation~\ref{eq:BasePrePostInteraction}: $\delta^{\mu} + \gamma > 0$ and $\gamma > 0$.\footnote{While $\delta^{\mu} < 0$ would fail our first hypothesis -- presumably because of a backfire effect of some sort -- our particular interest with the moderation result regards whether moderation occurs with respect to $PCE$, and whether the combined treatment effect of $PCE$ and $T$ is positive. Thus, for each DV, this hypothesis requires that both estimated quantities meet their respective conditions.}
To implement this test, having estimated \cref{eq:BasePrePostInteraction}, we simulate 2000 draws from the estimated sampling distributions of the coefficients, and then calculate what fraction of those draws meet the joint condition that $\delta^{\mu} + \gamma > 0$ and $\gamma > 0$, with the latter quantity denoted as $PCConsistent$. As we wish to adjust our inferences to reflect the multiple comparisons that we make across the five DVs, we calculate a $p$-value for our theoretical inference as $p_x = 1 - PCConsistent$, where $x$ indexes the DVs. We then adjust the set of five $p$-values, again using the \citet{BenjaminiHochberg1995} procedure.


\begin{table}[htp] 
	\input{\texdir/experiment_treatment_effects.tex}
	\caption{Treatment effects on Left-Right Ideological Beliefs}
	\footnotesize{\emph{Note:} Table presents effect estimates from each of the four ideological-link treatments as estimated from the OLS model described in \cref{eq:BasePrePost}. All dependent variables are standardised to have mean zero, standard deviation one. The multiple-comparison adjusted p-values for the F-test of joint significance of the treatment effects is presented for each model (\emph{F-test adj. p-value})}.
	\label{tab:experimentTreatmentEffects} 
\end{table}

\begin{table}[htp] 
	\input{\texdir/experiment_treatment_interaction_effects.tex}
	\caption{Interaction between ideological treatments and personal-experience prime}
	\footnotesize{\emph{Note:} Table presents effect estimates (as estimated from the OLS model described in \cref{eq:BasePrePostInteraction}) for a binary indicator capturing whether a respondent received one of the ideological-link treatments, and the interaction between that variable and a binary indicator which captures whether a respondent saw the personal-experience prime. All dependent variables are standardised to have mean zero, standard deviation one. P-values for the test of the moderation hypothesis, adjusted for multiple-comparisons, are presented for each model (\emph{Adj. p-value})}.
	\label{tab:experimentTreatmentInteractionEffects}
\end{table}


\begin{landscape}
	\begin{figure}[htp]
		\centering
		\vspace{2cm}
		\includegraphics[width=1.5\textwidth]{\plotsdir/treatment_effects.png}
		\caption{Effects of Ideological Treatments on Attitudes (coefficients and 95\% Confidence Intervals)}
		\label{fig:experiment_results}
		\emph{Note:}  Confidence intervals are not adjusted for multiple testing
	\end{figure}
\end{landscape}


\clearpage
\section{Ideological Rhetoric and Personal Crisis Experience}
\label{apdx:Exp:TripleInteraction}

In our main experimental analysis, we demonstrated that ideological rhetoric linking the COVID pandemic to broader considerations about redistribution and the role of government was largely ineffective at shifting voter attitudes. Even amongst those respondents who were encouraged to think about their own experiences of the crisis (i.e. those in the PCE priming condition), the effects of our ideological-link treatments were largely null. Moreover, in our observational analysis, we demonstrated that personal exposure to the crisis -- either in the form of lost employment or receipt of government financial support -- had little effect on voter attitudes.

However, it remains possible that ideological rhetoric linking the COVID-19 pandemic to broader issues relating to redistribution and the role of the state might have been especially effective for the set of voters who were more exposed to the financial dimensions of the crisis. That is, in terms of our experimental design, we might expect our ideological rhetoric treatments to have differential effects depending on whether respondents reported had been personally exposed to the crisis. That is, we might expect important interactions between reported personal experience and our ideological treatment texts. 

The interaction reported in the main text of the paper -- between ideological treatment and PCE prime -- shows that making the crisis more salient in people's minds does not affect the efficacy of the rhetorical treatments. In this section, we are interested in whether the ideological treatments were more or less effective at shifting the attitudes of respondents who reported having personal experience of the crisis. We also consider whether the differences in treatment effects for affected and non-affected individuals are smaller or larger for respondents who were primed to think about those experiences before answering our outcome questions. 

As described above, for all respondents we collected data on whether either they, or someone close to them, had lost their job, experienced a reduction in income, had a business fail, or whether they had used any of the forms of government support that we measured in our panel data.\footnote{These questions form the basis of our `personal COVID experience' (PCE) prime. However, when analysing the PCE prime in the main text, we are concerned only with whether respondents were asked these questions before answering our dependent variable questions (the primed group), or after answering our dependent variable questions (the non-primed group). Here, by contrast, we divide respondents on the basis of their responses to these question -- regardless on when they were asked them -- into groups indicating whether the respondent was personally affected by the crisis in some way.} In the analysis below, we use responses to those questions to code two binary indicators. The first measures whether a respondent experienced (or knew someone close to them who experienced) some kind of labour-market loss from COVID. The second measures whether a respondent received (or knew someone close to them who received) some kind of government support during COVID. 

We then reanalyse the experimental data with modified versions of equation \ref{eq:BasePrePostInteraction}, where we include three-way interactions between the ideological treatment variable, the PCE prime, and whether our respondents report having either received government support or experienced loss from COVID. This analysis allows us to compute conditional average treatment effects of the ideological-link treatments for respondents with different experiences of the crisis, and across the different conditions of our PCE prime. 

\afterpage{
	\begin{landscape}
		\begin{figure}[htp]
			\centering
			\vspace{2cm}
			\includegraphics[width=1.4\textwidth]{\plotsdir/treatment_triple_interaction_effects.pdf}
			\caption{Conditional treatment effects by personal experience and priming condition (coefficients and 95\% Confidence Intervals)}
			\label{fig:experiment_results_triple_interaction}
			
		\end{figure}
	\end{landscape}
}

We present the results of this analysis in figure \ref{fig:experiment_results_triple_interaction}. The figure shows that there is very little treatment-effect heterogeneity across respondents with different experiences of the crisis, and that these effects are largely insensitive to the PCE prime. In general, across all respondents, we largely find null effects of the ideological-link treatments and we rarely find significant differences between those who experienced COVID-related loss and those who did not, nor between those who received COVID-related support and those who did not. Further, there is no evidence that the differences in treatment effects are larger in the group of respondents who were primed to think about their experiences before answering the outcome questions.

Overall, these results are therefore consistent with the those presented in the main body of the paper in that they provide no evidence that politicians could have used different discourse during the pandemic to shift aggregate public opinion. Even voters who were personally affected by the crisis, and were primed to think about their experiences, were largely unpersuaded by ideological arguments linking the COVID pandemic to broader ideas about redistribution or the role of government in the economy. 

\clearpage
\section{Panel Attrition}
\label{apdx:Exp:Attrition}

In order for panel attrition to cause bias in the estimates that we present in the paper, it would have to be the case that the probability of remaining in the sample in the post-crisis-onset waves is different for respondents directly affected by the crisis (either in terms of job loss or receipt of government support during the pandemic) than for other respondents, and that between-wave retention is also related to the outcome variables we study. We are unable to directly assess whether there was differential attrition across our treatment and control groups as the treatment is only measured during the crisis period (i.e. we cannot observe which pre-crisis respondents would have gone on to receive furlough payments, Universal Credit, lost their jobs, and so on). 

We can, however, evaluate the degree to which the post-crisis-onset waves of the BES were subject to higher rates of attrition than the pre-crisis waves. Figure \ref{fig:between_wave_retention} shows the wave-on-wave retention for each survey wave in the BES. Wave 20, the post-crisis-onset wave, included 17,794 respondents from a total of 32,177 wave 19 respondents (the most proximate pre-crisis wave), which equates to a retention rate of 55.3\%. Two points are worthy of note. First, the retention in wave 20 is the lowest of any wave in the data. Second, between-wave retention has generally been trending downwards over time. This suggests that the low retention after the onset of the crisis was not a large aberration but rather reflects the continuation of a long-run trend.

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/aggregate_retention.pdf}
	\caption{Between-wave retention}
	\label{fig:between_wave_retention}
\end{figure}


\clearpage
%\section{Subjective risk of unemployment and receipt of government support by income group}
\section{Evidence on the Scale and Distinctiveness of the COVID-19 Economic and Policy Shock}
\label{apdx:SubjectiveUnemploymentRisk}

In this short section, we set out evidence regarding the nature of the economic shock that was inflicted by COVID-19 (on the UK), how it compares to more `normal' recessions, and the scale and distribution of government financial support that was deployed. To be sure, as in earlier recessions, the pandemic's economic effects were unequally distributed, yet the exposure of middle- and upper-income groups to economic hardship was exceptional. For example, \citet{Witteveen2020} estimates that, in the early months of the pandemic in the UK, economic hardship (defined as income loss) was roughly 2.5 times more likely among workers in the lowest income quintile as compared with those in the highest quintile. At the same time, exposure to hardship in the 3rd, 4th and 5th quintiles was quite significant, ranging from 20.1 percent in the 5th quintile to 36.4 percent in the 3rd quintile (p. 3). The age distribution of hardship tells a similar story: while workers aged 18-24 were roughly 1.5 times more likely than workers aged 45-54 to experience hardship, the percentage of 45-54 year-old workers exposed to hardship was a very substantial 33.3 percent (p. 7). By comparison, in the midst of the Great Recession in 2009, 18-24 year olds in the labour force were nearly 4 times as likely as those over 50 to face unemployment, while the unemployment rates of those aged 25-49 and those over 50 were just 6.0 and 4.4 percent \citep[R5]{BellBlanchflower2010}.

To reinforce the point about the degree to which more affluent people were affected by the economic shock, figure~\ref{fig:subjective_unemployment_risk_changes} shows the changes in respondents' subjective perceptions of unemployment risk by income group between waves~17 and~20 of the BES (the waves immediately before and after the onset of the pandemic). The x-axis in the figure measures the mean difference in the $riskUnemployment$ variable in standard deviations of that variable and the y-axis indicates the self-reported income group into which respondents fall. The figure reveals that the largest increases in perceived risk of unemployment were in the wealthiest groups of respondents (those earning \pounds 50,000 and over), while lower-earning respondents (with yearly gross income of less than \pounds 20,000) reported \emph{decreased} risk of unemployment after the onset of the crisis. 

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/change_in_riskunemployment_by_income.pdf}
	\caption{Subjective risk of unemployment, by income group}
	\label{fig:subjective_unemployment_risk_changes}
\end{figure}

Separate from the distribution of labor market risks, the pandemic also led to a huge and distributively-distinctive economic policy response. The furlough scheme covered 11.7~million employees and cost \pounds70~billion \autocite{PowellEtAl2021-12-23}, with a peak of 30\% of the UK workforce supported \autocite{ONS2020-06-18}, while the self-employment scheme cost a total of \pounds28 billion. The implementation and subsequent government reporting of the scheme makes it harder to be precise about how many people's income was supported. However, the respective waves of the scheme saw the following numbers of claims accepted: 2.6 million, 2.35 million, 2.2 million, 1.95 million, and 1.26 million \autocite{Seely2022-01-28}. Beyond these aggregate numbers, figure~\ref{fig:government_support_by_income} shows the fraction of respondents in pandemic waves of our panel who reported receiving support from the government during the crisis, for each income group. The figure reveals that the higher-earning income groups had greater exposure to the furlough scheme compared to lower-earning groups of respondents --  a pattern which is the opposite of that found for receipt of Universal Credit.

\begin{figure}[htp]
	\centering
	\vspace{2cm}
	\includegraphics[width=\textwidth]{\plotsdir/government_support_by_income.pdf}
	\caption{Receipt of government support, by income group}
	\label{fig:government_support_by_income}
\end{figure}


\clearpage
\printbibliography

\end{document}